Machine Learning-Based Predictive Models for Water Quality Index—An Analysis and Comparison

Water, being the most vital aspect of all living creatures and the environment, its availability and quality are paramount concerns for global sustainability. Water quality is a fundamental property that determines the suitability of water for a certain application based on its intended use and the required physical, chemical, and biological characteristics. Recent activities of humans has led to the concentrations of substances in natural water resources to exceed the ideal concentration values, resulting in a modification of the natural composition. This effect has led us to find an adequate solution for water quality index (WQI) calculation as well as the prediction of water quality classification (WQC) which is suitable for usage in everyday life, from publicly available data with ML-based predictive models being a driving force in the process. A number of preprocessing techniques were performed on the dataset, which includes the relevant imputation of some missing categorical and numeric data points, along with the use of unsupervised Non-Negative Matrix Factorization (NNMF) technique and balancing of the target data using SMOTE algorithm. Different machine learning models have been applied as a part of model training for WQC forecasting. The results obtained from these models were assessed, examined, and compared on the basis of their accuracy, recall, precision, as well as F1-score. Prediction results demonstrated that the Gradient Boosting Classifier achieved the best results (96.99%) for WQC prediction. Water Management resources can benefit greatly from this kind of promising research.